《拥有生物激励神经感知引擎的实时多目标识别处理器精选文档.ppt》由会员分享，可在线阅读，更多相关《拥有生物激励神经感知引擎的实时多目标识别处理器精选文档.ppt（13页珍藏版）》请在三一办公上搜索。
1、object recognition,First,various scale spaces are generated by a cascaded filtering for input video stream.Then,key-points are extracted among neighbor scale spaces by local maxima/minima search,and each of them is converted to a descriptor vector that describes the magnitude and orientation of it.L
2、ast,the final recognition is made by nearest neighbor matching with pre-defined object database that generally includes over ten thousands of object descriptor vectors.,Disadvantage of SIMD processors,their identical operations are not suitable for key-point or object level operations such as descri
3、ptor vector generation and database matching.the multi-core processor of exploits coarse-grained PEs and memory-centric network-on-chip(NoC)for task-level parallelism over data-level parallelism cannot provide enough computing power for real-time object recognition due to its data synchronization ov
4、erhead.,The papers work,Section II describes a visual perception based multi-object recognition algorithm in detail.Section III explains system architecture of the proposed processor.Detailed designs of each building block are explained in Section IV.Section V describes the architecture.proposed NoC
5、 communication The chip implementation and evaluation results follow in Section VI.,II.visual perception based multi-object recognition,A.Visual Perception Based Object Recognition Model,B.Overall Algorithm(注：(ROIs)regions-of-interest),III.system architecture,IV.building block design,A.Neural Perception Engine,B.SIMD Processor Unit,C.Decision Processor,V.proposed NoC communication,VI.Low-power techniques,Thanks for your attention,
